【作者】 潘浩；

【导师】 任玉波；

【作者基本信息】 燕山大学 ， 机械电子工程， 2025， 硕士

【摘要】 轴向柱塞泵是液压系统的核心动力元件,其寿命预测对保障液压系统的安全运行至关重要。然而,由于柱塞泵结构复杂、造价昂贵且工作环境恶劣,导致物理建模困难、数据稀缺与数据质量低,传统的基于单一失效物理或数据驱动的方法难以获得高精度的寿命预测。针对上述问题,本文提出了适用于轴向柱塞泵的机理—数据相容融合的剩余使用寿命预测方法,主要研究内容与结论如下:(1)针对柱塞泵物理退化模型建模困难的问题,在对柱塞泵失效机理分析的基础上,以柱塞泵滑靴副作为主要研究对象,构建了滑靴磨损-泄漏退化模型。首先,分析滑靴副的结构及工作原理,建立动力学方程,分析求解其受力与运动规律。然后,通过三点油膜法与滑靴副磨粒磨损模型,计算滑靴副油膜厚度并建立油膜润滑磨损模型。最后,基于压力流量特性建立了滑靴副磨损与泄漏相互表征的性能退化模型。(2)针对柱塞泵恶劣工况下数据质量低问题与多维长时序非线性退化数据的特点,提出一种考虑测量误差的卡尔曼滤波和改进粒子滤波融合算法对数据进行清洗,有效提升了数据的平滑效果和质量;采用多特征长短期记忆神经网络作为数据驱动算法,拟合退化特征与时间的关系曲线,提高了对时序非线性数据的预测效果。为进一步提升模型性能,通过麻雀搜索算法解决了神经网络中超参数选择与优化问题。(3)针对当前柱塞泵退化数据采集困难与单一模型预测效果不佳的问题,提出了一种物理-数据融合的剩余使用寿命预测方法。通过采集柱塞泵的结构与环境参数,基于滑靴副磨损性能退化模型迭代生成时序数据作为物理特征,验证其与预测目标的相关性并筛选高质量特征,提升了可用数据的数量;简化退化模型并引入未知参数,采用梯度下降法对参数优化求解,以补充预测步退化数据并构建物理信息损失函数,提高模型可靠性。实验结果显示融合模型在小样本与长期预测中相比单一预测方法具有更高的预测精度。更多还原

【Abstract】 Axial piston pumps are the core power components of hydraulic systems,and their life prediction is crucial for ensuring the safe operation of hydraulic systems.However,due to the complex structure,high cost,and harsh working environment of axial piston pumps,physical modeling is difficult,data is scarce and of low quality,and traditional methods based on a single failure physics or data-driven approach are difficult to achieve high-precision life prediction.To address these issues,this paper proposes a mechanism-data compatible fusion-based remaining useful life prediction method for axial piston pumps.The main research contents and conclusions are as follows:(1)To address the difficulty in modeling the physical degradation of axial piston pumps,based on the analysis of the failure mechanism of axial piston pumps,the slipper pair of the axial piston pump is taken as the main research object,and a slipper wear-leakage degradation model is constructed.Firstly,the structure and working principle of the slipper pair are analyzed,and the dynamic equation is established to analyze and solve the force and motion laws.Then,the oil film thickness of the slipper pair is calculated by the three-point oil film method and the slipper pair abrasive wear model,and an oil film lubrication wear model is established.Finally,based on the pressure-flow characteristics,a performance degradation model that mutually characterizes the wear and leakage of the slipper pair is established.(2)To address the problem of low data quality under harsh working conditions of axial piston pumps and the characteristics of multi-dimensional long-term nonlinear degradation data,a Kalman filter and improved particle filter fusion algorithm considering measurement errors is proposed to clean the data,effectively improving the smoothing effect and quality of the data.A multi-feature long short-term memory neural network is used as the data-driven algorithm to fit the relationship curve between degradation features and time,improving the prediction effect on time series nonlinear data.To further enhance the model performance,the sparrow search algorithm is used to solve the problem of hyperparameter selection and optimization in neural networks.(3)To address the current difficulties in collecting degradation data of axial piston pumps and the poor prediction effect of single models,a physical-data fusion-based remaining useful life prediction method is proposed.By collecting the structural and environmental parameters of the axial piston pump,the time series data is iteratively generated based on the slipper pair wear performance degradation model as physical features,verifying their correlation with the prediction target and screening high-quality features,increasing the amount of available data.The degradation model is simplified and unknown parameters are introduced,and the parameters are optimized and solved using the gradient descent method to supplement the degradation data at the prediction step and construct a physical information loss function,improving the model reliability.The experimental results show that the fusion model has higher prediction accuracy than single prediction methods in small sample and long-term prediction.更多还原